Database Vdisks spreading across controllers (EVA 8400) - Best Practice?

[mattwaddy]

Hi all we currently have our MS SQL Cluster (4 Nodes) setup with all Vdisks on Controller A. Now I would have thought we would be better to split the SQL Instances over the two controllers to better balance IO?

Ideally my thoughts are as follows;

We have a single VDisk from each of the SQL instances inside a DR group so in this case 3 Vidsks which are then replicated to the second site. As these are in a single DR group that means they must all be on the same controller. I'm thinking of splitting these up and spreading the instances over the two controllers as follows;

Instance 1 - Controller A

Instance 2 - Controller B

Instance 3 - Controller A

Quorum - Controller B

That would help balance IO over the EVA better i'm guessing, but I'm thinking of maybe going one step further.

Instance 1

SQL Data - Controller A

SQL Logs - Controller B

SQL TempDB - Controller B

SQL Backups Controller B

Instance 2

SQL Data - Controller B

SQL Logs - Controller A

SQL TempDB - Controller A

SQL Backups Controller A

Instance 3

SQL Data - Controller A

SQL Logs - Controller B

SQL TempDB - Controller B

SQL Backups Controller B

The thought being that load for each SQL instance would be spread over the two controllers rather than one, so when doing a SQL maintenance plan reads for data would be from one controller and writes to the backup vdisk would go to the second which should improve throughput and distribute load better. Does that make any sense at all? I'm hoping that someone can assist. We don't replicate sql data / tempdb / logs to the second site, we rely soley on Maintenance plan backups to the backup drive being replicated for recovery in a full disaster.

[Jan Soska]

Hello,

your setup looks well. It is good to spread load over both controllers.

In our enviroment - MS SQL + EVA6000 we did other interesting tests - we presented 2 luns form EVA (each form different controller) and did software raid 0 directly under windows (dynamic disks). Results were impresive. Both luns were from the same disk group, but read permormance raised from ~360MB/s to ~520MB/s thanks to involving both controllers into work.

But - this setup has some cons - you can not resize such lun easily, potencial risk of software raid failure.

Do some experiments on your own.

Jan